Semantic enhanced Top-k similarity search on weighted HIN

Abstract Similarity searches on heterogeneous information networks (HINs) have attracted wide attention from both industrial and academic areas in recent years; for example, they been used friend detection social collaborator recommendation coauthor networks. The structural the HIN can be captured by multiple metapaths, people usually utilize metapaths to design methods similarity search. rich semantics HINs are not only but also content stored nodes. However, of nodes was valued existing methods. Although some researchers recently considered types machine learning-based search, structure separately. To address this issue balancing influence flexibly process searching, we propose a double channel convolutional neural network model top-k which uses path instances as inputs generates embeddings based different metapaths. We an mechanism enhance differences each node. Another is combine Finally, importance evaluation function designed improve accuracy make more explainable. experimental results show that our search algorithm effectively support achieve higher performance than approaches.